"Noiseless" input stage

[Fancy Lime]

Hi everyone and a happy New Year!

I'm finally back after a crazy year with almost no time for diy electronics that whatsoever. I love the smell of solder in the morning. Smells like victory. Wait, no it doesn't. Smells like confusion and I need some pointers once again.

So I'm working on a modular, "proper" bass preamp / channel strip with all the bells and whistles: Compressor/limiter, drive stage(s), semi-parametric three or four band EQ, cab sim, headphone out, balanced line out, parallel and serial effects loops (possibly frequency selective). Most parts are pretty straight forward and only need some fine tuning to get the EQ frequencies etc. just right. My biggest problem right now is getting the the input stage as versatile as I need it and as noiseless as possible.

The difficulty with a bass preamp is, that it needs to be able to digest highly dynamic signals from low as well as high impedance sources (hundreds of ohms to hundreds of kilo ohms). I don't mind making the user select the right input or source impedance with a switch or dual inputs. The compressor can, if designed carefully, take care of the extreme dynamics if it goes on the very front before the gain stage. But that obviously hurts the signal to noise ratio because the is at the very least some additional series resistance before the first gain stage. So I decided the first gain stage goes on the very front before the compressor or anything else. For low impedance sources this is no problem: Non-inverting opamp stage with a low-noise BJT opamp (NE5532A, NJM2068, LM4562...) using +18/-9V (to handle the large input dynamic range) from an overclocked 1044 charge pump (like in the Klon Centaur). While this would technically have sufficient input impedance even for high impedance sources, the noise performance of these gets worse with higher source impedance.

So the question is: How to design the high impedance input? Three ideas:

1) I could of course just copy the low impedance stage but with a FET input opamp. OPA2134 is the only one that comes to mind as potentially useful. The good old TL072 is too noisy for my liking and things like the OPA627 are just plain unreasonable price wise. But is that really the best way to go?

2) Or can I get away with putting a simple source follower using a low noise JFET (2SK170, 2SK117, 2SK369...) in front of the BJT opamp and connect and bias it in such a way that its output sets the bias for the opamp, thus minimizing components that come in series before the gain stage. This will introduce extra noise but will it be worse than using a JFET opamp (who seem to be inherently noisier than the really low noise BJT ones if I interpret the datasheets at all correctly) under real life conditions.

3) Or would I better use a discrete solution using low noise components. The mu-amp/mini-booster/SRPP comes to mind, or a "discrete opamp" as in the Boss BD-2 Blues driver. I seem to remember R.G. posted a simple JFET-PNP buffer/booster here some years ago, that looked like an even simpler version of the BD-2 circuit, just cannot find it right now. Or even a simple common source stage with a JFET. Some sound coloration at this stage would not be considered a no-no unless it is really obvious or sounds unpleasant but the mentioned alternatives should probably be ok in that respect.

I am aware of some supposedly very low noise JFET cascode input stages but put them aside for the time being as too finicky (read that on some HiFi forum for whatever that is worth) and probably overkill for my purposes (not building a radio receiver, hopefully).

So: Can someone point me to a good source explaining how to calculate which of these would be lowest noise at a given source impedance and gain. I have so far not been able how to meaningfully compare the info in opamp datasheets to those in transistor datasheets (or even between opamp datasheets of different manufacturers or between FET and BJT input opamps) despite spending a substantial amount of time reading (overly simplified and partially contradictory) explanations.

Yes, I am aware that I might be overthinking this. I could just build a bunch of versions and try them and I would probably find that it all does not make that much of a difference either way. On the other hand, anything that comes before a compressor and one or more drive stages should be as dead quiet as reasonably possible if the whole thing is supposed to be useful. Or so my ocd says. Also it bugs me that I still don't understand how to properly design for low noise.

Thanks and cheers,
Andy

[dschwartz]

Interesting project!!
There are many factors that will affect noise..Even picking the quietest input stage will not warrantee a noiseless behaviour..
I'd go for the NE5532 solution, as it looks like the best cost effective and decent performance. Also I'd use low value resistors to minimize thermal noise.

Noise wise.. I'll bet the EQ section will give you more trouble..

[R.G.]

You're probably not overthinking it, as low noise front end design is a classical EE problem. Fortunately for your intentions, it's a well-plowed field as well. I suggest finding and absorbing one or more of the existing textbooks on low noise design.

You have correctly absorbed some of this already. There is a trade-off to be done based on source impedance. Noise comes from different mechanisms, and as both current noise and voltage noise. The boil-down on this leads to the different choices on front end design, and the differences in active device input impedance makes picking the "best" front end a thoughtful exercise.

The conventional wisdom is that you want low input impedance for low impedance sources, and high input impedance for high impedance sources, and all the interesting action happens at the edges of the middle impedance region, where clever and interesting new active devices can make a few db of difference under those special conditions. The wide difference in source impedance of a multi-henry pickup coil starts to frustrate these clever choices, as does the very wide signal bandwidth. The "best" choice at one frequency for low noise can be a mediocre or poor choice at another. This means, as you're realizing, an exercise in compromise, whether you trade off frequency response for noise performance, or accept a bit more noise for wider frequency response, or both. It's a tough business.

The general wisdom in low noise design is to limit your amplifier's bandwidth to what you really, really need to amplify and then to get a substantial chunk of amplification in the very first input stage to get the signal up out of the low-level noise in every following amplifier stage.

That first item is simple, upon a little thought. Noise exists and enters at all frequencies. If you have a wider input frequency window that you really have to have, you let more in. Of course, in audio design, you're faced with the contrary need to get a response from DC to daylight. The noise specs from devices that are seriously specified for noise is generally quoted as volts or amps per square root Hz. What this comes down to is Know Your Source.

The second is also simple - every active device makes noise internally, and adds it to the incoming noise. The best you can do is to make as few noise contributions from active devices as possible, and to fit those active devices to the incoming noise spectrum and bandwidth. That first device gives you enough current, voltage, or power gain to let the succeeding active devices' noise contribution be negligible compared to the now-amplified signal you feed them. It's a truism that you can't improve noise performance after the first amplifying stage, only keep what you get. Concentrate on the input stage.

And with all this in mind, calculate the numbers so you have some ability to find out objectively where you're standing. There are no noiseless processes. You're going to be dealing with compromise, and with all compromises, you're facing the question of how much is enough and now much is too much. How much noise is there from the source, and how much work, time, and cost in parts and complexity will you spend to minimize it. Compare your numbers against other pre-existing compromises and decide how hard you're going to push to get "better". For instance, you can make some very, very low-noise front ends for amplification stages by dnking the front end devices in liquid nitrogen. Are you willing to use a Dewar flask as an enclosure?  

What you have here is a GREAT opportunity to learn. Dig in, go find out how high are the shoulders of the giants we all stand upon. 

[Danich_ivanov]

I would start with power filtering and small caps in negative feedback loop, whether it is opamp or transistor, should help.

[Fancy Lime]

Hi everyone and thanks for the replies!

@dschwartz
Like R.G. pointed out, if we can lift the signal in the first stage way above the noise level produced by the opamps, then all following stages have a much diminished influence, even if there are quite many. So the EQ only needs to be "relatively quiet". The current plan is: One inverting opamp with one or two gyrator based mid bands followed by a two-band Baxandall stage. Unity gain with pots centered. Should not be horribly noisy with a NJM2068 or NE5532. At least some studio consoles seem to do it that way with some success.

@R.G. and Danich_ivanov
Well, yes: restricting the bandwidth is obviously the first step. I normally cap all stages at 6kHz for bass preamps, which helps a lot with the super annoying hiss. And everything above 6kHz does not occur useful to my ears in a bass sound, just annoying. However, I am trying to design something with some appeal beyond my own person and maybe even for use with other instruments, including acoustics. So I have already played with the idea to make the upper frequency limit switchable at least for the first stage. Maybe 6, 9 and 12kHz or something along those lines. When overdrive is wanted, one should choose 6kHz, for clean sounds 12kHz is OK. You get the idea, let the user make some of the trade-off decisions

I have also experimented a little with two-pole frequency limiting in the gain stage by using a NE5534 with an overly large external compensation cap to cripple the slew rate in addition to the feedback cap. It audibly reduces high frequency hiss with high impedance passive pickups. Unfortunately, both the TI and onsemi datasheets (which were the only ones I found for the NE5534) only specify the effect of 22pF and 47pF compensation caps (~70kHs and ~60kHz large signal response limit, respectively). My gut tells my that simple math applies here, so 470pF should correspond to a 6kHz limit, and the listening test seems to corroborate that. I have no scope but am beginning to fear I might need one, one of these days.

Another thing I have not mentioned in the original post is frequency dependence of noise. On the one hand, there is 1/frequency noise, which gets worse at low frequencies. On the other hand, this is an audio device, not a scientific detector of some kind, so we don't really care about absolute noise levels, only the noise we can actually hear. And since our ears are much more sensitive in the upper midrange, you know, frequencies of human speech, than at 20-100Hz, I'm guessing that "technically mediocre" noise levels at 1500Hz are a bigger problem in audio than "technically awful" noise levels at 40Hz. So there's that additional compromise to get right. To round things off nicely, impedance is just the worst. I mean, at low frequencies passive pickups have a nice few kilo ohms. But at the high end they easily get in the hundreds of k ohm range. So exactly where our ears are more sensitive, source resistance becomes the bigger problem. Fan. Tas. Tic.

I would be curious to see how professional studio hardware from the olden days solved that problem but have so far been unable to find any schematics of low-noise high-Z studio gear. May have to do with the fact that  do not really know what to search for. Mixer channels seem to mostly be content with low input-Z and pushing the impedance matching upstream. Which I could obviously also do, but it would feel like a bit of a cop-out.

About the idea with the liquid nitrogen: Don't tempt me! I do have easy access to liquid nitrogen and even liquid helium, which at 4K would be about 26db better in thermal noise than liquid nitrogen at 77K. But once we are down in the single digit kelvins, we might as well make use of superconductivity and make the whole thing from niobium Josephson junctions. But maybe, just maybe, there is such a thing as over-engineering a device that is supposed to produce a fat and dirty bass tone. Maybe. Or is there...

As to the GREAT opportunity to learn: Yepp, that is the primary objective of this endeavor, more than actually having a nice preamp in the end. Best case scenario: The discussion to get there and the resulting project may tech some others something, too. Man, I missed this forum.

Cheers,
Andy

[R.G.]

one or two gyrator based mid bands followed by a two-band Baxandall stage. Unity gain with pots centered. Should not be horribly noisy with a NJM2068 or NE5532. At least some studio consoles seem to do it that way with some success.

Just be aware that gyrators are noisier than real inductors of the same size. The active device noise gets amplified in the device trying to make V = L di/dt.

I have no scope but am beginning to fear I might need one, one of these days.

You're way past fumbling along without a scope. Just go do it. You'll be a lot happier with your electronics overall.

Another thing I have not mentioned in the original post is frequency dependence of noise. On the one hand, there is 1/frequency noise, which gets worse at low frequencies. On the other hand, this is an audio device, not a scientific detector of some kind, so we don't really care about absolute noise levels, only the noise we can actually hear. And since our ears are much more sensitive in the upper midrange, you know, frequencies of human speech, than at 20-100Hz, I'm guessing that "technically mediocre" noise levels at 1500Hz are a bigger problem in audio than "technically awful" noise levels at 40Hz. So there's that additional compromise to get right.

Fletcher-Munson aside, the ear does seem to get more hiss noise than midrange grind out of devices which are not particularly well designed for noise avoidance. Just from what exists as bad examples in the world, work on the highest frequency noise if you can't reduce it all. This is particularly true since most guitar-ish speakers have high frequency cutoffs in the range of 3kHz to 6khz themselves, so that's suppressing hiss as well. The active noise reduction schemes of the past, notably Dolby and a few others, were primarily hiss suppression systems.

About the idea with the liquid nitrogen: Don't tempt me! I do have easy access to liquid nitrogen and even liquid helium, which at 4K would be about 26db better in thermal noise than liquid nitrogen at 77K. But once we are down in the single digit kelvins, we might as well make use of superconductivity and make the whole thing from niobium Josephson junctions. But maybe, just maybe, there is such a thing as over-engineering a device that is supposed to produce a fat and dirty bass tone. Maybe. Or is there...

Good. Remember that a monstrous mind is a toy forever. I did see some yo-yo trying to hawk cryogenically treated vacuum tubes once. Laughed a bit at that.

As to the GREAT opportunity to learn: Yepp, that is the primary objective of this endeavor, more than actually having a nice preamp in the end.

Great! Right attitude!

[PRR]

> The compressor can, if designed carefully, take care of the extreme dynamics if it goes on the very front before the gain stage.

What compressor topology??

Supply logistics aside, some topologies can input huge signals, and you should put gain in front.

Some have relatively restricted dynamic range when used as limiters, and you get in a jam.

The possibilities are so numerous that I don't want to try to talk about "all" possibilities.

[Rob Strand]

1) I could of course just copy the low impedance stage but with a FET input opamp. OPA2134 is the only one that comes to mind as potentially useful. The good old TL072 is too noisy for my liking and things like the OPA627 are just plain unreasonable price wise. But is that really the best way to go?

For the first stage get the lowest noise JFET input opamp you want to put your money against.   Give it as much gain as you can within dynamic range restrictions; having a high-low gain inputs will help.  Once you do that the other stages aren't so critical (there might be a few specific areas that need attention).   Keep impedances sensibly low.   Don't use circuits that are inherently noisy or have a high noise gain.

FWIW, this type of semi-parametric is pretty good.  IIRC Eden amps use a similar thing,
https://proxy.duckduckgo.com/iu/?u=http%3A%2F%2Fi476.photobucket.com%2Falbums%2Frr130%2Fggeffects%2Fgg_simple_parametric.png&f=1

[amptramp]

Jump right in at the deep end:

http://ww1.microchip.com/downloads/en/AppNotes/01228a.pdf

[R.G.]

Yep, that looks like a good one. 

[Fancy Lime]

@R.G.

Good. Remember that a monstrous mind is a toy forever. I did see some yo-yo trying to hawk cryogenically treated vacuum tubes once. Laughed a bit at that.

You may laugh, I stand in awe of the ingenuity. If you can keep the tubes from cracking under the thermal stress you get stage fog bubbling right out of the guitar amp and get to use "NEW PATENTED CRYOTUBES FOR ULTIMATE SOUND" in the ads. That level of bullsh*t is an art form and we all know too many similar examples that do indeed sell very well. Lots of supposedly magic transistors, "warm sounding" resistors and "audiophile" power supply bypass caps to be found. My personal favorite: "audiophile power chords". Look ye here (and read the product description):
https://www.amazon.com/AudioQuest-NRG-Thunder-15A-1-0m/dp/B077J98CJ4/ref=pd_sbs_23_4?_encoding=UTF8&pd_rd_i=B077J98CJ4&pd_rd_r=b5a977e7-11a1-11e9-9239-3df688dacaec&pd_rd_w=K7EUe&pd_rd_wg=c35LA&pf_rd_p=7d5d9c3c-5e01-44ac-97fd-261afd40b865&pf_rd_r=8X8J35TJN9GE9TRXXD8B&psc=1&refRID=8X8J35TJN9GE9TRXXD8B


@Rob Strand
Nice, thanks! Simpler than a gyrator based solution and might be less noisy, will definitely try that one. Might be tricky to find a 100k rev log dual gang pot but since both parts are wired as variable resistors I might be able to fake it well enough with a linear 1M pot plus parallel 100k resistor from the top to the wiper. Interesting to have the control elements in a positive feedback loop, never seen it done like that. Looks a bit like a reverse baxandall type mid band.


@PRR

What compressor topology??

Supply logistics aside, some topologies can input huge signals, and you should put gain in front.

Some have relatively restricted dynamic range when used as limiters, and you get in a jam.

The possibilities are so numerous that I don't want to try to talk about "all" possibilities.

GOOD question. The major topology types I have so far considered:

1) OTA: Tend to not like large signal too much and seem to be difficult to get really quiet. So that one is probably out.

2) Diode bridge: Interesting but only works with very small signals, so no good here either.

3) JFET to ground in an L-pad attenuator: Nice and fast, but prone to distortion, especially with large signals. This one is still in the running but may prove impractical for my purpose. Studio compressors seem to mostly deal with the distortion problem by bootstrapping the ground end of the FET with a buffered signal in order to reduce voltage across the channel. However, I am a bit lost when it comes to the practical design details of this technique.

4) Vactrol: No problems with large signals but a bit on the slow side. Could go either in the feedback path of the input gain stage (think DOD 280A) or in an L-pad. The feedback loop solution has the disadvantage that the on and off resistances of the vactrol impose limits on the choice of feedback resistance to set the gain, if I want high compression rates and make the rate variable. So I would not be as free to optimize the gain stage for low noise as I would then putting the vactrol in an L-pad post gain stage. The major difficulty with vactrols is the response time. It certainly needs a full wave rectifier in the envelope detector, probably in form of a precision rectifier using a dual opamp. The rectifier then controls the current source for the vactrol. This design makes it very easy to control all possible parameters of the compressor independantly (although I'm going to set most parameters by design or trimpot and only have the necessary minimum of control by the user). However, this approach is never going to catch the more extreme transients one might get from slapping or a strong pick attack. So diodes in the feedback loop of the gain stage to chop of those are probably necessary (with a clipping indicator circuit of some kind). Keeping extreme transients as clean as possible is definitely not a priority for me in an instrument preamp.

5) Something entirely different that I have never seen done would be to use a Moog transistor ladder as a compressor. Should be possible by replacing the caps in the ladder with resistors and changing a few other things. But this would be a major undertaking to design and given the somewhat mysterious nature of the ladder filter, I am not overly confident that it will work satisfactorily at all.

So, probably vactrol, or maybe JFET.


@amptramp
Thanks Ron! I was unaware of that one. Seems a bit more to the point than the corresponding TI pamphlet, from what I have read so far.


Cheers and thanks for the input,
Andy

[PRR]

The Moog ladder is good for reducing small signals to very-small signals. Not what you want.

OTA, diode bridge, and Moog ladder are all the same thing: diodes. Maximum clean level across the diode about 20mV per diode.

The R+JFET pad gets you to maybe 50mV across the JFET. Output hiss level is the series resistor. For JFETs that can go down to 1K, and 30dB extreme reduction, you use a 30K resistor and get ~3uV hiss, perhaps 80dB output S/N. This is a fair choice.

If this was 1990 again, photoresistor would be THE way to go. "Slow" is reduced by NFB driver and choice of series resistor; also I believe in most musical applications a bit of under-compressed clipping on a sudden attack is benign, even good. However the restrictions on heavy metal sales crimp supply and quality.

To my mind the "today" solution is the THAT VCAs. Enormous dynamic range and precise characterization. I do not see the cost as a drawback; for 50 years we would have sold our car to get VCAs that good, but the THAT parts sell for half a tank of gas. Getting a linear limiter out of a log response does change the sidechain design. THAT has good notes, though with the attitude of "do it THIS way!", not a huge heap of alternative choices to "make it your own".

With hardly any exceptions (transformer coupled mikes into a tube limiter, maybe), I think you DO want a preamp. It may be high headroom, it probably should have variable gain (not a hi/lo pad in front). You pick a device with *both* low voltage hiss and low current hiss; selected JFET input opamp. You generally do NOT try to "match" low-Z sources (above 200 Ohms) but just gently sample the signal with a hi-Z input. (If a source needs loading as an expedient way to shape response, like 300r to tame an SM58, that is separate.)

And.... all this knobery reminds me of an old ART Tube Channel, preamp EQ limiter and tube stage in 1U. Meant perhaps as a singer's "sound in a box". I got it for XLR mikes but seem to recall an Instrument input.

[Fancy Lime]

Hi Paul,

VCAs! Completely forgot about those. I tried to wrap my head around a THAT4301 compressor but gave up quickly when I could not immediately figure out how to control attack, decay and knee softness. Then I played around with optical compressors a bit and found that I could do everything I wanted easily with those. But you are right of course, VCAs would "technically" be the better solution, especially because vactrol availability is indeed not great. I will investigate further. Not sure where you buy your THAT chips and gas but I get about 1/8 of a tank of gas for the price of a THAT4301, so cost is indeed not the inhibiting factor here.

The only real issue I have with the THAT4301 is that it's a power hungry little bugger. My original plan was to use a 9V external power supply boosted to +18/-9V using a TC1044s for the "headroom critical" stages only (i.e. before the limiter and for the line driver). That way I would have gotten away with a single charge pump. If I want to also supply a THAT4301 with +18/-9V, I need at least 3 TC1044s in parallel. At that point the 9V supply solution becomes a bit impractical. So if I go with the VCA compressor, I think I'm going to need either a massively more powerful charge pump (which I could not find from any of my usual suppliers), or a higher voltage power supply, probably 18V to 30V. Or an internal power supply but that would make the whole device bulkier than I'd like it to be.

Another possible type of compressor I forgot to mention is gain control by JFET, like in Jon Pattons Bearhug compressor. This should allow fairly large signals without subjecting the JFET to excessive voltage. But I have no experience with this, so I'd have to try. The same principle could also be applied to a non-inverting opamp stage (path from inverting input to ground via cap). I have never seen this done that way, wonder why.

With hardly any exceptions (transformer coupled mikes into a tube limiter, maybe), I think you DO want a preamp.

Not sure what you mean there. Of course I want a preamp. Do you mean preamp in the mixing console jargon, i.e. the very first gain stage? I certainly want that, too. The question is: Do I want the input to be able to deal with a a large variety of signals from different level and impedance sources (passive and active pickups, piezos, microphones, line signals) or am I content with having a very good input for middle to low impedance and delegate the impedance conversion for more demanding high-Z sources (passive pickups and piezos) to an additional external stage that should probably sit as close to the source as possible anyway. A phantom powered JFET buffer can always be put in the jack of a guitar cable. Just need to fit eh preamp input with the appropriate jack. Might be an interesting option for people who want to run 100ft cables without a separate buffer. Since I want to make the whole thing diy friendly and modular, this should probably included optionally anyway, come to think about it.

You pick a device with *both* low voltage hiss and low current hiss; selected JFET input opamp. You generally do NOT try to "match" low-Z sources (above 200 Ohms) but just gently sample the signal with a hi-Z input.

Well, it's not about matching low impedance as such but about getting the best noise performance from the input stage. Using a normal BJT opamp as a non-inverting amplifier provides plenty high enough input impedance for any except the most extreme sources. But while the noise performance of a BJT opamp is superior at low source impedance, JFET input opamps are a bit quieter at high source impedance but noisier at low impedance. Now if manufacturers would publish comparable noise specs, all would be well, I could just do the math from the datasheets. unfortunately, they don't often publish source impedance dependence of noise at all and when they do, it is not comparable. For example: the OPAx134 datasheet specifies voltage noise in nV/rootHz as a function of source resistance from 10 ohm to 10 Mohm, whereas the NJM2068 has (apparently weighted) "equivalent input noise voltage" (whatever that means) in µVrms and only up to 20 kohm. So not only do they use different measures, which may or may not be convertible into one another using additional information from the datasheet. The OPAx134 is missing the current noise and the NJM2068 has a weighting factor of undisclosed (?) nature. So at that point I wonder why the publish specs at all.

What I actually want from a datasheet is all noises as a function of source impedance as well as a function of frequency in a 3D plot. Yeah, I know, not happening. Also: that still does not tell me about the psycho-acoustic relevance of the respective noise. So I am reluctantly coming round to the practical solution of simply building a test rig for A-B testing different opamps with all relevant real life sources. I just need to build to exactly identical stages with matched resistors and such. Might me a while before I find the time since I am moving next week.

Cheers,
Andy

[amptramp]

In my former life, I used to do noise calculations all the time so nanovolts per root Hz and picoamps per root Hz are all old friends of mine.  I had one job where I was building an amplifier for a photoresistive detector with an output impedance of 50 ohms.  The best transistor for this turned out to be the 2N4405, an ordinary silicon PNP switching transistor.  I used a differential pair of these feeding an LM318 op amp and got a midband reading of 0.64 nanovolts per root Hz.  One thing that I felt good about was I calculated the LM318 noise to be 11 nV/root Hz since National Semiconductor had not published any noise value for it.  Later, National came out with an amended data sheet and the new noise value was 11 nV/root Hz.

Since you are looking for an amplifier for bass response, you need plots of the 1/f noise since that usually makes itself known below 1000 Hz.  An amplifier optimized for low-impedance sources will be noisy (if it works at all) with high-impedance sources and an amplifier optimized for high-impedances sources will be more noisy than an optimally-designed amplifier for low-impedance sources.

FET amplifiers are probably the only ones that will meet your needs for high-impedance inputs.  A FET op amp has the disadvantage that the differential stage has root 2 times the noise of an individual transistor, so a single discrete FET is best here as the input device.

To ensure your low-noise amplifier stays that way, add diode protection to the gates to stop voltage spikes or turn-on and turn-off transients from damaging the FET.  It may add a slight amount of noise but it will keep the FET from deteriorating and increasing the noise level over time.

[Rob Strand]

The best transistor for this turned out to be the 2N4405, an ordinary silicon PNP switching transistor.

Back in the days of finding low rbb' transistors.    IIRC there was a way you could see the 2N440x had low rbb' from the datasheet.   The LM394 used the parallel transistor idea.   The purpose built transistors these days have crazy low rbb's.  (I'm sure you know) low rbb might help phono preamps but it won't help guitar stuff.

[PRR]

Ron R. is your man for hiss.

The two basic mid-audio specs are hiss Voltage and hiss Current. With those two you (or Ron) can compute hiss for any impedance in hand.

BJTs used to trump FETs for low voltage. This has not been true for over a decade. It may turn true again as the fat FETs go out of production.

There is a low impedance where Vhiss hurts and a high impedance where Ihiss hurts. In a BJT at any specific current the spread between these two impedances is just beta hFE. However if the impedance is known you can pick a BJT current so the spread straddles the design impedance. Use 20uA for 10K, 1mA for 1K, etc.

For any stage-sensible interface, the JFET has "no" hiss current. Now it comes down to 1/Gm. Old small-area JFETs did not have Gm as high as BJTs. 2SK170 is a standby.... but today's first hit on Giggle is "Find Rare Components." (eBay will sell you as many fakes as you want.)

J310 seems to be available from reputable sources. At 10(!)mA it has 17,000uS Gm, which is 58 Ohms. Marginal for Ron's star-sight but pretty OK for 200 Ohm microphones, and surely anything with a 1/4" plug. (However the eN spec disagrees.....)

IMHO you want to use an opamp for the input. Discrete tests and calculations are at Ron's pay-grade, but not for DIY unless you are very inclined to numbers with too many zeros and specsheets that don't spec. Opamp salesfolk know to make their stuff easy to spec and use.

[PRR]

> low rbb might help phono preamps

Moving COIL phono: essential. (At least for needle-up hiss.)

Dynamic microphones, you want pretty low rbb because it is rarely lots-less than mike impedance. The THAT parts seem to be the standard fix today.

For moving-Magnet phono needles, rbb is pointless (all but the worst Si BJTs will be plenty low).

[Rob Strand]

For moving-Magnet phono needles, rbb is pointless (all but the worst Si BJTs will be plenty low).

In the early days, it was pretty common to see rbb's of 400 ohms which did make a difference.   The main problem being people used "low noise" transistors but they weren't low noise for low impedance sources.
Over time people ratted out the devices which were low rbb's.   The ultra-low rbb's you get today are so low they don't add any improvement.

[Fancy Lime]

Hi there!

@Ron

FET amplifiers are probably the only ones that will meet your needs for high-impedance inputs.  A FET op amp has the disadvantage that the differential stage has root 2 times the noise of an individual transistor, so a single discrete FET is best here as the input device.

So, that means a single common source JFET stage with a decent amount of gain, right? Maybe even avoid variable resistors in the JFET stage and fix the gain to, say, 10 and then make the following stage an inverting opamp with gains from 1/10 to 5 (or whatever)? It seems logical to me that this must be quieter (theoretically) than an opamp with its many transistors if we assume perfect DC from the power rails. As soon as there is a realistic (for what is still basically a stompbox) power-supply attached, how does the single JFET fare. Or rather: how would I make sure it stays a quiet as it can?

Also: are there any decent JFETs in current production? I'm guessing through hole is out of the question for anything other than NOS. But even for SMD devices I have difficulty finding low noise types that are still in production. Although most of that problem comes from not knowing what IS in production, from sources other than semi-rumurous forum posts.

To ensure your low-noise amplifier stays that way, add diode protection to the gates to stop voltage spikes or turn-on and turn-off transients from damaging the FET.  It may add a slight amount of noise but it will keep the FET from deteriorating and increasing the noise level over time.

Do you mean a pair of antiparallel diodes or back-to-back zeners from gate to whatever the bias voltage is. So, parallel to the bias resistor? What voltage drop do I want? Also: In many designs, JFET inputs have a series resistor (typically 1k to 10k) for static discharge protection (?), I guess. That obviously does not improve the noise, so I guess I want to keep it low, if it is necessary at all in addition to the diode protection. How do I choose the optimum value?

Thanks,
Andy

[Fancy Lime]

So it seems apart from Ron the majority opinion favors a simple non-inverting JFET-input opamp stage for the input gain stage. For illustration purposes, I drew one (assuming split power supply):



Couple questions:
1) What opamp? The only one I have tested to any serious extent in such a position is the TL072. And that one is definitely too noisy. It is noisier than a NE5532 even when fed from a high impedance source. The OPAx134 series seems fairly popular and I have an OPA2134 lying around, so I'll definitely give that a try. Does anyone have other suggestions for very low noise JFET opamps? Looking at datasheets, the super-expensive ones like the OPA627 don't really seem to have an advantage here despite nominally better noise voltages because above 10k source resistance, the noise is dominated by the resistance anyway. But maybe someone has an underappreciated favorite.

2) What about the series Resistor R1? Adds noise, no question. Necessary for protection of the delicate JFET input or not? What is the smallest value that would be recommended here? Adding 1k to a typical passive guitar output impedance should ot make too much difference. 100k on the other hand...

3) Zeners: I added Z1,2 for JFET protection as per Rons suggestion (if I understood that right). What values to choose here? The OPAx134 datasheet specifies a mix input voltage of supply +0.7V on both ends but surly it would be advisable to set it to much less than the absolute maximum. Z3,4 are to protect the opamp from clipping and will be chosen depending on power supply voltage and  datasheet. I also want to add an indicator circuit that kicks in as soon as the diodes conduct and keeps it glowing for a fraction of a second even after they stop. Have to think about that a bit, can't be too difficult with a transistor as a current source and a cap to keep the charge a bit.

4) Gain control: If it is placed as it is in the drawing, then the variable R3 forms a variable high pass with C3, meaning that the bandwidth decreases as gain increases. We might want to keep the bandwidth constant, though. So maybe move the gain control to the path to ground (R4)? Then it only affects the lower end of the bandwidth, which probably wont bother us as much in terms of noise?

I almost miss the days when I soldered my first bazz fuss and just marveled at the fact that it made any sound at all without bothering why exactly it did what. Growing up is hard godsdamnit!

Cheers,
Andy